Additive scales demonstrated positive skew, so the natural logarithms of each were taken to normalize their distributions.The alpha coefficients ranged from 0.73 – 0.80 across the eight years.Studies have also found that patterns of marijuana use and depressive symptoms vary by several demographic characteristics. Thus, the following control variables were included in the model: family income and family structure.In addition,by the end of the observation period many respondents were young adults and a modest proportion had married, cohabited, or had a child. Since these life changes may affect the propensity to use marijuana or experience depressive symptoms, each year included dummy variables that indicated whether or not the respondent was married or cohabiting, or had a resident child. Finally, age was included in each model. Since preliminary analyses suggested that age had a quadratic association with depressive symptoms, both age and age-squared were included in the models.
Table 1 provides information about the distributions of all the variables used in the analysis. The standard deviations were decomposed into within-year and between-year components. Note that, for several of the variables, there was a similar degree of variability within and between years. The hypotheses were designed to compare four potential associations between marijuana use and depressive symptoms among young people. One of the hypotheses claimed that there are reciprocal associations between the two outcomes.Assessing reciprocal associations is complicated since the error terms of equations designed to predict two outcomes are not independent, thus failing to satisfy a key statistical assumption of most regression models. Another complication is that unobserved factors may confound any association between the two outcomes.In order to obviate each of these potential limitations, a fixed-effects regression model was estimated using an instrumental variables approach designed for longitudinal data. Fixed-effects regression adjusts for unobserved invariant factors and thus can reveal causal patterns among time-varying explanatory and outcome variables. The explanatory variables were each measured in the year prior to the outcome variables to better establish their causal ordering.For example, in the equation designed to predict marijuana use, depressive symptoms, stressful life events, self-esteem, self-efficacy and the other covariates were measured at time t − 1.
The results of the fixed-effects models are presented in Table 2.The coefficients represent within-person changes in the outcome variables for each one-unit change in the explanatory variables. In general, the results provided support for hypothesis 1 and failed to support the other three hypotheses. Eachone-unit increase in marijuana use was associated with a 1.75 increase in depressive symptoms relative to an individual’s average depressive symptoms score.Although stressful life events were also associated with changes in marijuana use and depressive symptoms, they did not attenuate the effects of marijuana use on depressive symptoms . In addition, there was no evidence from the empirical model that depressive symptoms led to changes in subsequent marijuana use, thus indicating that neither hypothesis 2 nor 3 was supported.The other results were consistent with previous research . Higher levels of self-esteem and self-efficacy were associated with fewer depressive symptoms. Peer substance use led to increases in marijuana use overtime. It was also associated with lower levels of depression, which may be indicative of how interpersonal associations, even with deviant peers, provide support that protects youth from issues of depression.Fixed-effects regression models assume that the effects of time-invariant factors,which are not directly estimated by these models, do not change over time. For example, variables such as gender and race/ethnicity are assumed to have the same influence on, say, depression and marijuana use regardless of the year or the age at which they are measured.
One way to test this assumption is to introduce interaction terms between time-invariant factors and time-varying explanatory variables. Since research has suggested that the effects of stress and family relations on marijuana use and depression may differ by gender,the models were extended to include interaction terms of each . The results of this robustness check indicated that no interactions attained statistical significance, thus further justifying the use of the fixed-effects approach. This study was designed to examine the association between marijuana use and depressive symptoms across a period of the life course, adolescence and young adulthood, when significant changes are occurring among individuals. Previous studies suggested that marijuana use and depressive symptoms may be associated in a unidirectional or a bidirectional manner; or their association may be explained by stressful life experiences or other factors that are unobserved inmost research studies.